Chaotic optical communication decryption framework based on the conv-transformer model

Chaotic optical communication is considered a promising approach to enhancing physical-layer security for optical communications. However, existing neural network-based chaotic communication systems suffer from high sensitivity to chaos synchronization and limited decryption accuracy. This paper proposes a chaotic optical communication decryption framework based on the conv-transformer model to address these issues. The core introduces a conv-transformer-based deep model, effectively integrating the Transformer's global attention mechanism with the convolutional layers' local perceptive capability, significantly enhancing decryption performance and accuracy. Additionally, a learnable differential connection is designed to embed the chaos synchronization process within the neural network architecture, simplifying the training and deployment process. The model achieves 100% decryption accuracy on a million-scale dataset. Compared with traditional methods, the proposed model demonstrates stronger adaptability and stability across various chaos system parameter configurations and channel conditions, while maintaining high sensitivity to key-related parameters. It effectively enhances decryption performance without compromising system security. Experimental results validate its feasibility and application potential in practical optical communication scenarios, providing a high-precision, easily deployable, and low-cost solution for secure optical communication.